2026-07-04

claiming strong GLM 5.2 serving results on AMD MI355X versus Nvidia Blackwell/B200

another data point that AMD inference economics may be getting competitive for open-weight serving, especially if the software stack is tuned hard.

Original source

Reading Log X Post Historical Backfill AI Infra

What it is: X post from wafer claiming strong GLM 5.2 serving results on AMD MI355X versus Nvidia Blackwell/B200.

Newsletter angle: “AMD is no longer just the cheap alternative” framing, with the real story likely in compiler/kernel/serving-stack optimization rather than raw silicon alone.

Retrieval note: extracted via FXTwitter API; inspected attached image for visible metrics; full reply-thread write-up not retrieved.